A hybrid framework for mining spatial characteristics of sewer defects from inspection databases

Certain sewer deterioration modeling approaches, represent the condition of a pipe as the sum of individual defect severities, leading to the loss of spatial information about defects in the pipe. This paper proposes a hybrid framework for incorporating spatial information of defects into the analysis of sewer pipe condition. The first component of the framework is called Defect Cluster Analysis (DCA) and it seeks to identify defect clusters (i.e. areas with multiple defects in proximity) and quantify their severity. The second component of the framework is called Defect Co-Occurrence Mining and it attempts to identify groups of defects, which occur simultaneously in pipes. The framework was evaluated on data from 7193 inspections of sewers in the US. Validation of the results by subject matter experts indicates that the proposed framework enables a fine-grained analysis of pipe condition and could be instrumental in rehabilitation decision-making.